PSGSL: A Probabilistic Framework Integrating Semantic Scene Understanding and Gas Sensing for Gas Source Localization

TL;DR

The paper addresses gas source localization under limited olfactory data by fusing olfactory measurements with semantic scene information using a probabilistic framework. It introduces a modular formulation that decouples olfaction and semantics, yielding a product posterior $p(s|g,z,α) ∝ p(s|g) p(s|z,α)$ and supports a per-cell map with a single-source XOR constraint $α$. Key contributions include an ontology-based semantic model linking object classes to gas sources, extensions to voxelized 3D maps and probabilistic gas classification, and an information-gain objective to guide navigation. Validation in simulation demonstrates that incorporating semantic cues speeds up localization and improves accuracy, highlighting the practical value of vision-augmented GSL and offering a flexible, extensible framework for integrating semantic information with existing GSL techniques.

Abstract

Semantic scene understanding allows a robotic agent to reason about problems in complex ways, using information from multiple and varied sensors to make deductions about a particular matter. As a result, this form of intelligent robotics is capable of performing more complex tasks and achieving more precise results than simpler approaches based on single data sources. However, these improved capabilities come at the cost of higher complexity, both computational and in terms of design. Due to the increased design complexity, formal approaches for exploiting semantic understanding become necessary. We present here a probabilistic formulation for integrating semantic knowledge into the process of gas source localization (GSL). The problem of GSL poses many unsolved challenges, and proposed solutions need to contend with the constraining limitations of sensing hardware. By exploiting semantic scene understanding, we can leverage other sources of information, such as vision, to improve the estimation of the source location. We show how our formulation can be applied to pre-existing GSL algorithms and the effect that including semantic data has on the produced estimations of the location of the source.

Figures (6)

• Figure 1: Bayesian networks modeling the conditional dependencies of the random variables involved in the problem. Network A models the source location through a single variable $S$ with a discrete domain, while network B uses a set of binary variables with an added XOR constraint to achieve desirable d-separation properties.
• Figure 2: Modified Bayesian network that accounts for dynamic, probabilistic gas classification.
• Figure 3: Voxelized semantic map reconstructed with Voxeland voxeland during the search process. The colored voxels denote a specific semantic category.
• Figure 4: Evolution of the error in the estimated source position over the course of the search operating with different amounts of semantic information. It can be observed that the version with no semantic data is slower to identify the correct area, and achieves lower precision in the final declaration.
• Figure 5: Progression of the source probability distribution through the search process. It can be observed that with semantics (room categories and camera observations) the localization converges more quickly to the correct position and achieves higher precision on the source declaration. The dashed line indicates the path followed by the robot.